MediaMeter Knowledge Base

Statistics for journalists

Mean vs. median, percentage points, margins of error, base rates, and correlation vs. causation — the numbers reporters most often get wrong, and how to get them right.

David Randall called innumeracy a fatal weakness for journalists. Numbers in news copy face the same discipline as quotes: they must be accurate, in context, and not made to say more than they mean.

The recurring traps

Mean vs. median

The mean (average) is the sum divided by the count; the median is the middle value. Income, wealth, and house prices skew — a few extreme values drag the mean upward. “Average income rose” can be true while typical citizens got poorer. Always ask: which average?

Percentage change vs. percentage points

If unemployment rises from 5% to 6%, that is a one percentage point increase — and a 20% relative increase. Confusing the two is among the most common factual errors in news copy. The accurate phrasing is always the points plus, where useful, the base (“up from 5%”).

Base rates

“Crime doubled in the village” from 1 case to 2 is technically true and practically meaningless. Any change needs the base figure and, for small counts, scepticism. Per-capita and per-100,000 rates make comparisons fair across different-sized populations.

Margin of error and samples

A 1,000-person poll typically carries a ±3-point margin at 95% confidence. A candidate “leading” 48% to 46% is statistically tied — say so. Check who was sampled, when, how the question was worded, and whether a subgroup’s small n makes a breakdown meaningless.

Correlation is not causation

“Cities with more bike lanes have more cyclists” does not prove lanes cause cycling (the reverse, weather, wealth, or demographics might explain it). Use careful verbs — “associated with,” “linked to” — and note alternative explanations. If the source itself makes a causal claim, attribute the claim: “the authors argue…”

Relative vs. absolute risk

“Doubles your risk of X” can mean a jump from 0.01% to 0.02%. Absolute figures keep readers calibrated: “two in ten thousand instead of one.”

The working checklist

  • Name the numerator and denominator — what is being measured out of what?
  • Give the base, the change, and the timeframe.
  • Prefer official primary sources (statistics offices, WHO, Eurostat, central banks); attribute all derived figures.
  • Check the units — billions vs. millions, per year vs. cumulative, nominal vs. inflation-adjusted.
  • Beware survivorship bias, self-selected samples, and press-release statistics with a sponsor.
  • Round honestly: “about 60%” beats false precision like “59.7%” for survey findings.
  • When a number feels surprising, treat that as a prompt to double-check — surprising numbers are disproportionately wrong.

MediaMeter context: numeric accuracy and statistical context sit inside the Accuracy and Context dimensions of the assessment specification — including the check that a figure’s claimed change uses the right base and unit.

Last reviewed: October 20, 2026. Part of the MediaMeter knowledge base — editorial reference material, not product documentation.

More from the knowledge base

The ten elements of journalism

Kovach and Rosenstiel's principles for journalism that serves citizens: truth, verification, independence, and the watchdog role — and why they matter more online, not less.

Journalism ethics codes compared

SPJ, NUJ, IFJ, AFP, Reuters, the German Pressekodex and the UK's IPSO code: what the major ethics frameworks require, where they agree, and where they differ.

News values and story strength

How editors decide what is news: Galtung and Ruge's classic factors, David Randall's eight values, and why "say stories" sit at the bottom of the scale.